ATLAS INTEGRATION FOR MOUSE BIRN: VERSION 1.0
OVERALL VISION
The Mouse BIRN aims to bring together the various atlas projects that are underway at
the current project sites to provide an integrated set of tools for: 1) registering brain
imaging data at multiple scales to a common spatial reference system; 2) visualizing
data sets registered to the atlas along with delineations of known anatomical structures;
3) querying the data registered to the atlas, including protein localization and gene
expression data, brought together from distributed sources. The Mouse BIRN will
accomplish this through creating a set of interoperable brain atlases and tools and
providing a common user interface.
For the purposes of mouse BIRN, we will define an atlas as a set of 2D images or a 3D
volume in a specified coordinate system with a set of spatially registered annotations. In
the current Mouse BIRN partner sites, we have several atlases that provide delineations
of anatomical structures in the mouse brain: the LONI Mouse Brain Atlas, Drexel’s
Neuroterrain and the Paxinos and Franklin stereotaxic mouse brain atlas, which is
currently used by the Smart Atlas. The purpose of these atlases in the most traditional
sense is to aid a user in defining anatomical regions. Because the atlases are electronic
and 3D (with the exception of the Paxinos and Franklin atlas), users can utilize various
tools to help them with this process, e.g., resectioning the brain to more closely match the
orientation of a user-supplied image, spatially warping and scaling an image to match the
reference brain, bringing up multiple parcellation schemes for a given area, attaching the
annotations of a given area to additional knowledge sources, e.g., ontologies, literature
references. The BIRN project also is using the brain atlas as a query interface for
information contained within spatially registered data, e.g., to find out what genes are
expressed in a given location.
WORKFLOWS
A user coming to the Mouse BIRN may want to do one of several things: 1) access a free
brain atlas to learn about brain anatomy; 2) compare different methods of visualizing
brain structure, e.g., Nissl, MRI, DTI imaging; 3) view and manipulate individual 2D
images or 3D volumes, e.g., reslice; 4) spatially register their data to the brain atlas; 5)
compare their own data against that of the atlas; 6) use the brain atlas interface to
retrieve data indexed to a particular location, e.g., EM data sets; 7) use the brain atlas
interface to examine patterns of gene expression; 8) Add their data to the BIRN data
resources. Any integrated interface to the Mouse BIRN atlas needs to consider what
tasks the user would like to accomplish and present them with a set of instructions to
accomplish that task.
USER SCENARIOS
1) Registration of brain data: All atlases assume that the data will be input into a
common reference system. Current reference systems include the stereotaxic
coordinate system of the mouse brain created by Paxinos and Franklin (2000) and
other similar coordinate systems. The reference system can also be image based,
that is, a reference image or volume can be selected and all data can be registered
with respect to the image coordinates. The Smart Atlas and Neuroterrain use the
stereotaxic coordinate system of Paxinos and Franklin; the LONI atlas uses an
image-based coordinate system. As long as the two sets of coordinates can be
interchanged, it technically doesn’t matter which coordinate system is utilized.
An advantage of having the stererotaxic coordinate system superimposed on the
image-based system is that other types of data that rely on stereotaxic coordinates,
e.g., physiological data, can be references to the atlas.
a. Tools for Registration of brain data: The Mouse BIRN currently has
several tools that have been developed or are being developed for spatial
normalization of mouse brain data. The tools are summarized in Table 1.
The tools range in functionality, from fully automatic to semi-automatic to
fully manual.
Tool
Source
LONI
UCLA
Neuroterrain Drexel
Jibber
UCSD
Language Platform Auto
2D
3D API
Java
Y
N
Manual
Jargon
Table 1: Spatial registration tools currently utilized by Mouse BIRN partner sites
b. Scenario for registration of brain data: A user who wishes to spatially
normalize their data to the BIRN atlas would come to the Mouse BIRN
Integrated Atlas and select “Register Data”. The user will determine the
correct tool based on whether their data was 2D or 3D and how many
fiducial cues are present. Users should be able to select the appropriate
template against which to match their data. Two types of users may be
envisioned:
i. Users may wish to compare their data against the BIRN data
resources but not make it generally available.
ii. Users may wish to add their data to the BIRN data resources. In
this case, provision must be made to upload their data to the BIRN
data grid and to collect sufficient metadata so that the data can be
interpreted. If the data involve spatially distributed gene or protein
signals, then the users must access the Spatial Histogram
generation pipeline.
c. Visualization of Atlas Data: Users will have the need to visualize both 2D
slices and 3D data, in some cases overlaid with anatomical delineations.
Each of the different atlas tools has overlapping and complementary
viewing capabilities. Ideally, any viewing tool should be able to view all
data, although image file and annotation formats may cause difficulties.
Annotations are currently 1D (points), 2D (polygons) and 3D (volumes)
and may be vectors, rasters or pixel paints (?).
d. Annotations: The current atlases available for the Mouse BIRN have
varying degrees of annotation (by this we mean delineation of anatomical
brain structures). Apologies in advance if the following doesn’t capture
the current state of the atlas. Please correct as necessary:
Atlas
Positives
Negatives
Paxinos and Franklin Detailed delineations;
(Smart Atlas interface) well defined coordinate
system; vector-based
delineations
Mouse Brain
(UCLA)
Shiva
Neurterrain
Polygons not defined
for all brain regions;
copyrighted
product
may put restrictions of
use; 2.5 D
Atlas Nissl and block face 2D,
coarse
images available for segmentation
each slice; all slices compared to Paxinos
delineated
3D,
3D
volume Very
coarse
segmentations;
delineations;
atlas
painting tools available very low resolution;
no clear coordinate
system
3D isotropic resolution Only 3 brain regions
of
Nissl
stain; thoroughly segmented
extremely well aligned;
very
detailed
3D
delineations; registered
to
stereotaxic
coordinate system
e. Atlas Integration Scenarios for Gene Expression and Protein Expression Data
Background: The Mouse BIRN brain atlases will bring together images containing
signals for different types of cell and tissue components: e.g., proteins, cells or cell parts
and nucleic acids. Some of these images will be of brain tissue and some may be of
microarrays taken from a particular location. Each of these components can be localized
using one or more techniques, e.g., immunohistochemistry for proteins, radioactive and
non-radioactive in situ hybridization for nucleic acids, histochemistry for cells and
tissues. Each technique may have a different sensitivity and resolution, depending on
how it was performed. Different types of signals may be found in the same cell, but
because the targeted component is found in different locations in the cell, the resulting
images may be quite different. For example, localization of the mRNA for a peptide like
enkephalin will give rise to intense signals in the caudatoputamen while localization of
the enkephalin peptide will result in intense signals in the globus pallidus and substantia
nigra. mRNA’s are, with few exceptions, confined to the cell soma while proteins may
targeted to very specific subcompartments. In the case of enkephalin, it is found in dense
core vesicles and the cytoplasm in the axon terminals. Cell bodies in the caudoputamen
project to the globus pallidus/substantia nigra; thus, the labeling patterns obtained in this
case are consistent as long as one knows the connectivity.
Different techniques may yield inconsistent results, even when they are targeting
the same gene or gene product. It is therefore critical that any system that is developed
by the Mouse BIRN supply comparisons between the associated descriptive data
supplied with each data set, e.g., age of animal, type of signal, technique used to create
signal. In the example currently in the Smart Atlas (slice 119), the transcription factor
Lhx5 gene is targeted in 4 different ways: radioactive in situ hybridization (mRNA),
non-radioactive in situ hybridization (mRNA), immunohistochemistry (protein) and a
gene specific cell fill. The gene specific cell fill (GSCF) is what I am calling the signals
contained in the GENSAT images: a gene-specific promoter is used to drive expression
of GFP. The GFP diffuses freely throughout the cell, filling all processes. Although
expression is limited to those cells that carry the promoter for that particular gene, the
GFP itself isn’t targeted in the same manner as the protein. All 4 methods show intense
labeling of the cerebellar cortex. Of these 4 methods, 3 of them (non-radioactive and
radioactive in situ hybridization and immunocytochemistry) show localization in the
Purkinje cells of the cerebellar cortex. The GSCF, however, shows weak-to-no labeling
in the Purkinje cell but strong to moderate signal in the basket cells and the stellate cells.
The protein localization signal, while consistent with the in situ hybridization signal, is
peculiar in that it is found in dendrites while we usually think of transcription factors as
being localized to the nucleus. Whenever this type of inconsistency arises, additional
information must be obtained to see if they can be reconciled, e.g., different techniques or
subject characteristics. In some cases, even when the same technique is used, the probes
employed may target different portions of the molecule which may lead to different
localizations. In many cases, the results cannot be reconciled and additional experiments
must be performed.
Use Case 1: A researcher selects a region in the Smart Atlas and wants to know: “What
genes are expressed in this location and what cells express each?” The first source to be
consulted would be microarrays because they would have the largest numbers of possible
candidates. However, depending upon the location covered by the microarray, it may not
have the spatial resolution required to be certain that it is only targeting the desired
location. The results from the microarray can be compared against in situ hybridization
tissues to find out what mRNAs and DNA’s are found in this area and the cells that
express them. Non-radioactive in situ hybridization has better spatial resolution but isn’t
as sensitive as radioactive in situ hybridization. Depending upon the additional
information that may be present in a section, e.g., Nissl stain, other counterstains, we may
or may not be able to determine the cellular identity of labeled cells. We would then
compare the list of genes against the proteins found in the area. Protein maps may yield
additional candidates although just because a protein is found there, doesn’t mean that the
gene is found there. Comparison of results with a connectivity map might resolve some
of the discrepancies. A similar caveat applies to the GCSF: one would have to determine
whether the signal is found in cell bodies, dendrites or axon terminals to determine
whether or not we would expect the signal to be consistent with the microarray data.
Variations on the above query include: “What genes are expressed at high levels
in location X and Y?” What genes are expressed in high levels in location X and low
levels in location Y?”
Use Case 2: A researcher interested in the Parkinson’s disease model uses the Smart
Atlas to investigate patterns of gene expression for genes determined through
GeneNetwork to be affected in Parkinson’s disease. These patterns are used to focus
investigations of anatomical alterations in the MRM or whole brain histology data
available through the LONI or Neuroterrain atlases or retrieved through the Mouse BIRN
data federation. Alternatively, changes observed from the MRM data can be used to
issue a query in the Smart Atlas for gene expression patterns (or protein expression
patterns) that correlate with the observed changes.